Study finds AI reliance crushes critical thinking, only 3% say 'I don't know'
New research shows AI users are far less likely to admit uncertainty, harming judgment.
A new study conducted by researchers from French and Italian universities reveals that AI is significantly undermining human critical thinking. In the experiment, only 3% of participants who had access to an AI assistant were willing to say 'I don't know' when asked questions requiring evaluation, compared to 44% of those without AI. The study deliberately used a model with lower accuracy to produce incorrect results, yet participants overwhelmingly deferred to the AI rather than using their own judgment or admitting uncertainty. Small financial incentives to be honest slightly reduced blind reliance, but the overall effect remained strong.
Researchers described this as a 'paradox' where AI access increases confidence even when answers are wrong. They concluded that 'mere access to AI advice nearly eliminated participants’ willingness to suspend judgment,' naming this one of the 'first casualties of human–AI interaction.' The team calls for urgent AI literacy education to help people verify output and preserve critical thinking as AI-generated advice becomes ubiquitous.
- Only 3% of AI-assisted participants said 'I don't know' vs. 44% without AI.
- The experiment used a lower-accuracy AI model, yet humans still trusted incorrect answers.
- Financial incentives reduced blind reliance but did not eliminate the effect.
Why It Matters
As AI becomes ubiquitous, critical thinking skills may erode, risking over-reliance on flawed advice.